Simulated prosthetic vision : object recognition and localization approach

Currently, there is no efficient treatment against eye diseases such as retinitis pigmentosa (RP) or a ge related macular degeneration (ARMD). They affect mi llions of people worldwide and may result in blindness after a few years of evolution [1]. Withi n the last decades, a large collection of assistive devices have been designed to assist people with vi sual mpairment and enhance their autonomy. However, assistive devices didn’t prove to be very useful in complex tasks such as navigation or object recognition and grasping. In the meantime so me research groups aim to restore vision via neural interfaces, and have designed varied neuropr osthesis that elicit perceptible spots of light (phosphenes). Today, the most advanced prostheses a re implanted in the retina, and are based on two different strategies. In the first one, an image is captured from a micro camera mounted on a pair of goggles, then transmitted to a video processor and finally converted into electric pulses by an electrode array implant. In the second one, an arra y of light sensors implanted in the retina directly converts the incoming light into electrical signals [2]. Both devices have been designed to restore an image of the whole field of view through a point-to -p int display of phosphenes (sometimes called “scoreboard” approach). The first clinical trials w ith arrays of 60 electrodes are encouraging: implanted blind subjects are able to perceive simpl e visual stimuli in a highly contrasted and controlled environment. However, more complex visuo -m tor tasks, such as grasping an object among other objects, are still very limited with th ese neuroprosthesis.